Product
Metrics & KPI Framework
Proposed North Star metric, business/product/AI/ops KPIs, and an explicit instrumentation-gap inventory.
Source: docs/product-management/12-product-metrics-kpi-framework.md
HotelMind AI — Metrics & KPI Framework
Scope: Business, product, AI/ML, and operational KPIs. No target/benchmark values are asserted as measured facts unless labeled otherwise — a target with no live customers to validate it is a planning assumption, not a benchmark. Metrics with no current instrumentation are flagged as such rather than fabricated.
Proposed North Star Metric
Weekly AI Recommendation Acceptance Rate — the share of pricing/restaurant/staffing recommendations approved (versus rejected or overridden) by an authorized user in a given week.
Why appropriate: it directly reflects whether users trust and act on the platform's core value proposition — AI-informed operational decisions — rather than a vanity usage metric.
Feasibility: not currently instrumented. The guardrail/recommendation-approval table exists in the database, but no dedicated analytics pipeline aggregates acceptance rate into a reportable metric today. This is a near-term product-analytics gap, not a technical blocker.
Business KPIs
| Metric | Data Source | Status |
|---|---|---|
| Revenue, Occupancy, ADR, RevPAR | Warehouse fact tables | Computable; no baseline/target set — no live customers |
| Food Waste | Real order data | Not currently instrumented — no real order data exists |
| Guest Satisfaction (CSAT proxy) | Review-analysis pipeline | Directional only — synthetic review data today |
| Staff Utilization | Real attendance data | Not currently instrumented — no real attendance data |
Product KPIs
| Metric | Status |
|---|---|
| Active Users | Computable from auth logs; no adoption baseline pre-launch |
| Dashboard Module Adoption | Not currently instrumented |
| Feature Adoption (per domain) | Not currently instrumented |
| Recommendation Acceptance Rate | See North Star metric — guardrail data exists, not yet aggregated |
| AI Assistant Usage | Not currently instrumented |
| Alert Engagement | Not currently instrumented |
| Report Usage | Not currently instrumented |
AI/ML KPIs
| Metric | Status |
|---|---|
| Occupancy Forecast MAPE | Measured: 72–120% — currently poor; should not be presented as high-confidence |
| Restaurant Demand Accuracy | Not measurable against real data yet — synthetic validation only |
| Staffing Demand Accuracy | Not measurable against real data yet — synthetic validation only |
| Churn Model Precision/Recall | Invalid today — label leakage; do not report current metrics as valid |
| Model Drift | Instrumented (KL-divergence / PSI), engineering-only |
| Recommendation Confidence Calibration | Not currently instrumented |
| AI Assistant Response Quality | Not currently instrumented — no human-eval pipeline |
Operational KPIs
| Metric | Status |
|---|---|
| Booking Processing (success / attempted) | Instrumented via backend logs |
| Room Utilization | Computable from warehouse |
| Restaurant Demand / Staff Planning Accuracy | See AI/ML KPIs above |
Instrumentation Gap Summary
The majority of Product KPIs and several AI/ML KPIs are currently not instrumented. HotelMind AI has the underlying data — guardrail approvals, model predictions — but lacks a dedicated product-analytics/telemetry layer to aggregate it into reportable metrics. This is flagged here rather than fabricated, and building that instrumentation layer is a near-term recommended action.